Grey Wolf Optimizer Enhances Adaptive Atrous Spatial Pyramid Pooling for Efficient Multi-Scale Feature Selection in Medical Image Segmentation

This study proposes enhancing DeepLabV3+ by incorporating the Grey Wolf Optimizer (GWO) for adaptive channel selection into the Atrous Spatial Pyramid Pooling (ASPP) module. The proposed enhancement helps the model prioritize informative features, improving segmentation accuracy in complex scenarios like brain tumor detection. The modification in the Atrous Spatial Pyramid Pooling module as suggested in this study with incorporation of adaptive channels allows context-based feature capture at multiple scales necessary in order to achieve accurate segmentation. Analyses conducted on MRI dataset have proved that the addition of GWO improves the mean Intersection over Union (mIoU) score of the model to 75.8±0.8%, which is a remarkable improvement over the baseline score of 73.1±0.8% obtained by the baseline DeepLabV3+ model. In addition, the model achieves a greater Dice score of 82.7±1.3% as well as an accuracy rate of 99.300± 0.047%, outperforming rival models like FPN and FCN-ResAlexNet. The approach utilized in GWO facilitates the selection of highly relevant channels with minimal redundancy, thereby enhancing feature representation. The model also tackles essential challenges that relate to classification imbalance, hence maintaining a level of stability in a variety of circumstances. The addition of a two-stage convolution methodology with global context incorporation with images greatly improves the competence level of the model, therefore making it a viable alternative in real-time medical images. The study highlights possibilities in deeper models’ methodologies in order to attain improved competence in difficult segmentation.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.